Consider renaming TraceT

[jonsterling]

The general pattern is to have a monad signature given in Control.Monad.Sig.Class and then a concrete structure, which is the "canonical" implementation of that signature, in Control.Monad.Trans.SigT.

But TraceT is not the canonical instance of MonadTrace, it is a particular instance which combines tracing with error handling in a useful way. The canonical instance would be a minimal wrapper around StateT, probably.

To avoid confusing people, the TraceT type might be better named something else.

[ghost]

How about using a free monad? Wouldn't that give you a good notion of canonicity and a simpler structure (referring to #8)?

[jonsterling]

Interesting idea! I'm certainly open to it.

[ghost]

I'll take a shot at it. I see you're already pulling in Codensity :)

[jonsterling]

Great! :)

Yeah, I did my best to make it so you can just program against the abstract type, which means that we have a lot of flexibility in improving the internal representation. Thanks!

[ghost]

How's this? https://gist.github.com/darinmorrison/9f6d757b1cc07865cc41

Didn't do it as a monad transformer yet but that shouldn't be hard. Could use Free instead too.

[jonsterling]

This is pretty cool! I haven't used operational or free very much, are there any performance issues I need to worry about?

[ghost]

Operational's Program should right associate the binds already so you don't need Codensity. This is similar to the Rust thing I'm doing. If you do it CPS style like with Free, you will want to use that though. Actually you don't quite even need Codensity IIRC. Kmett has a "smaller" monad transformer that still works for that but I forget the details just now.

[ghost]

Also, I would be pretty surprised if the Program version isn't faster than the transformer stack version, even with Codensity or similar tricks.

[jonsterling]

Heh, that's neat. Yeah, sounds good!

[ghost]

Okay. I can add the missing stuff and put together a pull request in a bit.